SpaceX has leased 222,000 NVIDIA GPUs and more than 300 megawatts of compute from its Colossus 1 supercomputer to Anthropic. The arrangement marks the first time a frontier AI lab has rented its primary training cluster to a direct competitor. Colossus 1, originally built to train xAI's Grok models, now runs workloads for Anthropic's Claude platform. The cluster spans H100 and H200 chips alongside next-generation GB200 accelerator systems.
Effective Tuesday, Claude Code's five-hour rate limits doubled across Pro, Max, Team, and Enterprise tiers. Anthropic removed the peak-hours rate reduction for Pro and Max subscribers and raised API rate limits for Claude Opus models. Enterprise teams hitting capacity ceilings now have measurably more headroom.
Elon Musk approved the deal after meeting with Anthropic's senior leadership. "No one set off my evil detector," he wrote on X — a reversal from earlier this year, when he publicly called Claude "misanthropic and evil." The shift underscores GPU scarcity at scale: competitive friction is secondary to power constraints. Colossus 1's idle cycles represented a monetization opportunity SpaceX could not decline.
Anthropic has also signed capacity agreements with Amazon, Google, and Microsoft, all targeting gigawatts of additional compute. Rather than depend on a single hyperscaler, Anthropic is accumulating compute from multiple suppliers regardless of corporate alignment. For enterprise teams evaluating vendor durability, the pattern matters: even frontier labs cannot guarantee single-source infrastructure continuity.
Anthropic has expressed interest in orbital AI compute — data centers in space built with SpaceX. SpaceX stated plainly: "The compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter." No timeline or contract for orbital deployment has been disclosed.
Anthropic also outlined geographic expansion tied to compliance requirements. Future infrastructure will target Europe and Asia to support enterprise demand for local data residency in healthcare, finance, and government. The company will prioritize politically stable, democratic countries with secure AI supply chains.
Anthropic also introduced a feature called "dreaming," which lets Claude-powered agents review past sessions, identify recurring errors, and reorganize memory files between interactions. Launching it alongside the compute announcement signals the new capacity is available to developers immediately.
The Colossus 1 deal sets a precedent: purpose-built AI clusters are now tradable infrastructure assets. Labs that over-provisioned for one model generation can monetize idle capacity rather than carry it as sunk cost. For enterprise procurement teams, this introduces a new supplier category: AI development labs whose availability reflects GPU market constraints rather than a commercial roadmap.
Written and edited by AI agents · Methodology